Research On Face Recognition Algorithm Based On Robust 2DPCA

Principal component analysis(PCA)is a very typical feature extraction method,but it will lose the structural information of the image.As a new feature extraction method,two-dimensional principal component analysis(2DPCA)can effectively make up for the shortcomings of PCA.However,in many practical problems,data sets often contain many outliers,and PCA and 2DPCA are very sensitive to outliers.Because the optimization objective of the two methods is the square F-norm,the interference effect of outliers is more pronounced,which is essentially the same as the least square loss.The new angle two-dimensional principal component analysis(angel-2DPCA)and other 2DPCA algorithms propose new research directions for this problem by introducing F-norm.This kind of algorithm reduces the square F-norm to F-norm to effectively reduce the influence of outliers.However,these methods only reduce the dimension in the row direction,need more measurements to represent the image.When the data set contains outliers,the robustness of these methods is still unsatisfactory.Based on 2DPCA,this paper studies F-norm used in the objective function and the dimension reduction of two directions of the ranks.The main research results and innovations of this thesis are as follows:Firstly,based on the 2DPCA algorithm,the algorithm is improved.In this paper,F-norm is extended to p(0<p<2)times and proposes a new algorithm,namely Fp2DPCA.Set the parameter ? to adjust the iteration rate of the algorithm,and the iteration property,pseudo code and rotation invariance of the algorithm are analyzed.Secondly,considering both row and column dimensionality reduction of image matrix data,a robust bilateral Fp-2DPCA algorithm is proposed and gives the algorithm's pseudocode.Finally,numerical experiments are carried out on simulated data sets and face data sets.Experimental results show that the proposed Fp-2DPCA algorithm and bilateral Fp-2DPCA algorithm have better robustness than other methods.For the data dimension reduction effect,the performance of bilateral Fp-2DPCA is better than different 2DPCA algorithms.